Representative Engagements

Case 1 - Fair Lending Staff Augmentation following a Top 3 Bank Merger/Acquisition

A leading bank requested staff support following the departure of seasoned compliance officers following an announcement of one of the largest acquisitions for the year. An existing inventory of statistical models were analyzed to determine the fair lending risk associated with the models. Conclusions following the review led to over 75% of the models receiving independent variable revisions. Additionally, analyses were performed on the following functions:

  • Marketing Propensity Models

  • Discriminant Bias

  • Proportional Parity

  • Underwriting Disparity

Institution Scope:
International Bank & Large Bank, 600+ billion in combined assets.

Case 2 - Fair Servicing Reviews for two Large Mortgage Servicers

Following the COVID-19 pandemic, two large mortgage servicers requested a fair servicing review to identify potential fair servicing risks following the implementation of the CARES Act. Advanced statistical modeling was custom built for each servicer and risk scorecards were generated to identify fair lending risk focal points. Focal points were further researched using a comparative file process to identify potential disparities by exit disposition following the forbearance period. Analyses focused on the following areas to determine the potential impact on minority borrowers:

  • Forbearance/Loss Mitigation Plan Disposition

  • Forbearance Plan Exit Dispositions

  • Loss Mitigation Waterfall Qualitative Review

Institution Scope:
Mortgage Servicers, 75+ billion in combined assets.

Case 3 - Fair Lending Performance Analysis & Program Review in Numerous Consultative Engagements

Fair lending analytics have been performed on numerous institutions (listed below) as the most common form detective controls for risk management. These analyses generally come in two forms, performance (quantitative) and program (qualitative) reviews. In the past, a wide variety of available lending data was analyzed for fair lending performance for numerous clients ranging from large to small institutions across multiple financial sectors. The most common data type used is an institution’s HMDA LAR for mortgage records. The following below are examples of data science analytics performed for institutions with available LAR data to determine how minorities were treated relative to a similarly situated non-minority borrower:

  • Underwriting Disparity (Approvals v. Denials)

    • Cross Tabulation with Chi-square for identifying variable relationships

    • t-Test (both one- and two-tailed) for statistical significance between two group means (minorities and non-minorities)

    • Logistical Regression was used to identify the decision to approve or deny a minority borrower based on credit worthiness factors (e.g., LTV, DTI, Credit Score)

  • Pricing Disparity (APRs Variances)

    • Linear Regression was used to identify the APR (dependent variable) relates to credit worthiness factors serving as independent variables

    • Ordinary Least Squares (OLS) was used to build a fitment line and determine how statistically significant credit worthiness factors were on pricing

  • Levels of Service

    • Processing Times - days from application date to action code are assessed on absolute disparity to determine minority borrower focal points

    • Fallout Index - withdrawn and incomplete applications are assessed based on absolute disparity to determine minority borrower focal points

  • Redlining

    • Performance indices were established for a defined geography, either Reasonably Expected Market Area (REMA) or bank CRA Assessment Area, when applicable

    • Peer groups were created for 50-200% of the institution’s lending activity for each defined geography and compared to the institution’s lending performance with respect to Majority-Minority Census Tracts (MMCT) and Low to Moderate Income Tracts (LMI)

    • Select institutions opted to perform advanced peer group comparisons for Top 3 Peers, All Lenders, and Strategic Peers for each geography

  • Steering

    • Fixed v. Adjustable Rate Mortgage (ARM) and FHA v. Conventional Mortgage analyses is performed to determine if a borrower was solicitated to purchase an ARM loan when they would have qualified for fixed-rate mortgage of the same amount.

Recent developments have stressed going “beyond HMDA” with lending data analytics. As a result, the following types of unconventional data have been utilized for top performing lenders across the country via a Bayesian Improved Surname and Geocoding (BISG) proxy methodology:

  • Consumer Non-real Estate (Personal Loans, Auto Loans, RV Loans, Solar Panel Lending)

  • Credit Cards: both secured and unsecured

  • Business Lending: SBA loans, Small to Midsize Enterprise, Commercial Lending

  • Complaints Data: volume analysis by disposition/protected class, geographical distribution by MMCT and LMI.

  • Limited English Proficiency (LEP) Borrowers: call handling (abandonment rate, hold times, Interactive Voice Response performance)

Institution Scope:
Mortgage Companies ($1 billion to $200+ billion), including one of the top 10 mortgage companies by asset size.

Credit Unions ($1 billion to 19+ billion in assets), includes three of the top 25 credit unions by asset size.

Community Bank (< $10 billion in assets)

Mid-size Bank ($10-50+ billion in assets)

Large Bank ($50+ billion in assets)

Case 4 - Fair Lending Model Validation - Credit Reporting Agency Model Analytics for a Leading FinTech

In the contemporary financial landscape, ensuring the precision and fairness of credit reporting agency models has become paramount. To this end, advanced analytics were conducted on a top fintech's credit reporting agency model. Here's an overview of the rigorous analysis undertaken:

  • Model Validation and Performance Testing:

    • Performance Metrics Evaluation: Determining model accuracy, sensitivity, specificity, and area under the ROC curve.

    • Stress Testing: Evaluating model robustness under varying scenarios, especially those involving economic downturns.

    • Overfitting Checks: Employing techniques such as cross-validation to ensure the model's generalizability.

  • Fair Lending Disparate Impact Analysis:

    • Underwriting Analysis: Investigating discrepancies between approvals and denials, especially concerning minority applicants.

    • Variable Relationships: Utilizing Cross Tabulation with Chi-square tests.

    • Statistical Significance Testing: Using t-Tests (both one- and two-tailed) to compare group means.

    • Decision Analytics: Employing Logistic Regression to discern the factors influencing approval or denial decisions for minority borrowers based on credit factors such as credit history, outstanding debts, and income.

  • Geocoding Analysis:

    • Redlining Indicators: Establishing performance metrics for defined geographies, focusing primarily on geocoded data concerning the institution's operational areas.

    • Comparative Analysis: Instituting peer group evaluations based on geocoded data to discern lending patterns concerning Majority-Minority Census Tracts (MMCT) and Low to Moderate Income Tracts (LMI).

    • Geographical Performance: Advanced peer group analyses involving Top Peers, All Lenders, and Strategic Peers were meticulously executed.

  • Advanced Analytics:

    • Beyond Traditional Metrics: Incorporating unconventional data sources for a holistic model analysis using the Bayesian Improved Surname and Geocoding (BISG) proxy method.

    • Diverse Portfolio Analysis: Delving into areas like:

    • Personal Credit Products: e.g., Overdraft facilities, Payday loans, Student loans.

    • FinTech Specific Products: Peer-to-peer lending, Robo-advisory recommendations, Crowdfunding projects.

    • Customer Feedback: Analyzing complaint data, focusing on disposition, geographical patterns, and minority representation.

    • Language Proficiency: Evaluating service metrics for Limited English Proficiency (LEP) users, such as chatbot efficiency, document translation accuracy, and digital platform navigability.

Institutional Scope:

  • Digital-Only FinTech Platforms: Ranging from startups to those valued at $50+ billion.

  • Hybrid FinTech Institutions: Those combining traditional banking services with innovative financial solutions, assets ranging from $10 billion to $150+ billion.

  • Partnerships & Collaborations: Working closely with large banks, credit unions, and other financial entities to ensure integration efficacy and model reliability.